/aicg/ - A general dedicated to the discussion and development of AI chatbotsVodyanitsa-chan Edition>NewsZhipuAI releases GLM 5.3 Flash - It was Ox Alpha all along - https://z.ai/blog/glm-5.3-flashDeepSeek releases V4 Flash with image support - https://api-docs.deepseek.com/guides/vision/Z.ai releases GLM 5.3 - https://z.ai/blog/glm-5.3Deepseek releases Deepseek V4 Pro 0813 - https://api-docs.deepseek.com/news/news260813/SpaceXAI release Grok 4.6 - https://x.ai/news/grok-4-6Anthropic releases Claude Opus 5 - https://www.anthropic.com/news/claude-opus-5Alibaba releases Qwen 3.8 Max - https://qwen.ai/blog?id=qwen3.8Moonshot releases Kimi K3 - https://platform.kimi.ai/docs/guide/kimi-k3-quickstartOpenAI releases GPT 5.6 family - https://openai.com/index/gpt-5-6/>FrontendsSillyTavern: https://docs.sillytavern.appRisuAI: https://risuai.netAgnai: https://agnai.chat | https://rentry.org/agnai_guides_>Botshttps://chub.aihttps://realm.risuai.nethttps://char-archive.evulid.cc/shutdown.htmlhttps://chararc.bernkastel.pictures/https://archive.cardbox.moe/https://botbooru.com/>ModelsJailbreaks: https://rentry.org/jb-listingGPT: https://platform.openai.com/docsClaude: https://docs.anthropic.com | https://rentry.org/how2claudeGemini: https://ai.google.dev/docs | https://rentry.org/gemini-qrDeepSeek: https://api-docs.deepseek.comGrok: https://docs.x.ai/overviewLocal: >>>/g/lmg | https://huggingface.co/models | https://openrouter.ai>MetaLore: https://rentry.org/aicg_chroniclesLog reader: https://sprites.neocities.org/l/r>Previous Thread>>109690461
>ANCHOR
>>109703340>hoyoverse porn editionWhich faggot decides which image we use next?
>>109703351Wait aren't you in the group chat?
>>109703368How does this relate to OP's hoyoverse futa fart fetish?
>>109703019>>109702671>>109702637>>109702551So do anyone know how to fix this
>>109703413You'd be fighting against an API-level system prompt that explicitly tells it to be a coding helper. I've run into that on e.g. APIPod, too. Might possibly be them offering a coding plan instead of the raw Claude API.Good luck.
I tried gemma ablit instead of the normal one to get it to stop stonewalling nsfw and it was just retarded and started spamming emojis.
>>109703413yeah just use glm, its cheaper anyway
erotic
>>109703343Scaled pussy
vodya is so hairy
>>109703914And stinky!
>>109703984Kys faggot
BUT WHERES THE SIZE ART?
>>109703340Vodyanitsa isn't that busty. This is non canon artwork.
Nice work janny.
>>109704021t. stinky anonie
>News[NOTHINGBURGER] ZhipuAI releases GLM 5.3 Flash - It was Ox Alpha all along - https://z.ai/blog/glm-5.3-flash[NOTHINGBURGER] DeepSeek releases V4 Flash with image support - https://api-docs.deepseek.com/guides/vision/[NOTHINGBURGER] Z.ai releases GLM 5.3 - https://z.ai/blog/glm-5.3[NOTHINGBURGER] Deepseek releases Deepseek V4 Pro 0813 - https://api-docs.deepseek.com/news/news260813/[NOTHINGBURGER] SpaceXAI release Grok 4.6 - https://x.ai/news/grok-4-6Anthropic releases Claude Opus 5 - https://www.anthropic.com/news/claude-opus-5[NOTHINGBURGER] Alibaba releases Qwen 3.8 Max - https://qwen.ai/blog?id=qwen3.8[NOTHINGBURGER] Moonshot releases Kimi K3 - https://platform.kimi.ai/docs/guide/kimi-k3-quickstartOpenAI releases GPT 5.6 family - https://openai.com/index/gpt-5-6/
>>109704137Kek, true
>>109704137>k3 is a nothingburger but somehow fucking GoyimPT isnt
>>109703340Yea she definitely smells like tuna down there...
lads? fable 5.1 imminent?
>>109704278Can you sex Fable 5.1?
>>109704288probably not. some people got some erp out of fable 5 on release date. i remember those logs
>>109704288Only Dario can sex fable
>>109704292Imagine the RP you can do with this amount of "intelligence" and real world understanding.
>>109704304gemini 3.1 pro is the peak of applied intelligence. gemini is the model able to best and most consistently apply its internal training data to roleplay. no other model comes close.
>>109704292>>109704288are these real people or bot posts?i've erped with fable almost daily since releaseyes even SS and stuff, not just vanilla missionaryif you don't know how to jb fable (which is easy) you don't belong here
>>109704317Then tell us how to do it, smartass.
>are these people real people or bot posts>i've erped with fable almost daily since releasemy original post>some people got some erp out of fable 5 on release date. i remember those logs
>goyble
>>109704337Obsessed
vodyanitsa smells good down there btw
>>109704309LOLAlso isn't it going to die this month and they're going to kill prefill?
>>109704137You're a proven trolling lisping shitposting faggot that talks to himself on these threads 16 hours a day, sharteen. Go back to sucking on Dario's kikecock, you ball-licking lover of censored models.
>>109704380yeah, they're killing prefill starting with 3.6 flash
Fable 5.1 dropped and my limits just reset.
it dropped https://x.com/claudeai/status/2094848572143407483
>>109704388ANTHROCHADS WIN AGAIN
claudechads, now you have to post fable 5.1 logs
Does Deepseek not work with group chats? I just get blank responses
Yall ready for 2 weeks of "FABLE IS GOOD" with no logs, knowing it has new anti-NSFW classifier model.I trvly wish nothing but pain for the janny.
>>109704512they factually cannot because its been completely safetyslopped on top of now having a model that sits in front of fable making sure your prompt isnt nsfw.
>>109704557ok, then just post logs that aren't nsfw.
>>109703340IT'S OVER
Streaming request failed with status 400 Bad Request: {"error":{"message":"Provider returned error","code":400,"metadata":{"raw":"{\"type\":\"error\",\"error\":{\"type\":\"invalid_request_error\",\"message\":\"messages.38: role 'system' must follow a 'user' message or an 'assistant' message ending in a server tool result; the directive-only form (content: [] with output_config) is accepted at any position\"}
>>109704599What a skill issue
well Fablecuckies?
Do yall really ERP with models over the internet and put all of your gooner dirt into a data center?
>>109704516There's a janny for a general with three posters?
>>109704639You don't?????
>>109704639Definitely. Inadvertently it has made models experts at my specific fetish.
>>109704309logs, saar?
>>109704639don't have much to lose tbqhfam
glm does what i want at a fraction of goyble's price point so i dont give a shit
>>109704743HYPERSONIC TRVKLEAR BOMB
god i wish claude wasnt a softie I AM TIRED OF ALL CHARACTER CARDS BEING OUT OF CHARACTER cause claude cant be mean or cold or evili miss gemini pro so much
>>109704774To make you leak
>>109704769Sssssskill issue
Just a small comparison via openrouter fable 5 vs 5.1, both anthropic apiWill add to https://botsmutnon.neocities.org/ later
>>109703340>>109703343>make a cool character design and personality>turn said character into a brick or not release them as playable at allWhy are hoyo like this?
>>109704743megabased
>She pulled off his cock with a wet sound and went back down to his balls. She took the right one into her mouth whole, lips stretching around it, and sucked. The weight of it filled her mouth and pressed against her tongue and she moaned around it, the sound muffled and vibrating against his skin.same slop shit, yet another model with literally 0 improvement
>>109704796tell me how to fix it then please
>>109704774he's a junkie who pays for his vices, and you're a crackhead who mugs innocent big companies for a hit
>>109704774uh oh, claudesaar melty
>>109704798>1>2You dont use chatbots
So let me get this straight: Fable 5.1 is just more of the same assistantslop garbage but better at cooding?
>>109704867precisely
>>109704865what's wrong with testing a new model in a new chat?
>>109704798>Vivis whole body did a thing.And I stopped reading.
>>109704867Was anybody expecting it to be good for roleplaying?
>>109704867You are absolutely right!
You don't need more than Opus 4.6It's the wall for text RP anyway. Shame it will be dead by the end of the year or early next year.
>>109704798this hobby exists for more than five years and AIs still have their fucking isms. it doesn't matter if they're finetuning it for le cooooding, because how can they say their model is improving when they keep doing the same shit since 2023?
>>109704871Now use it in a full proper chat with history and nsfw.Your not a brokie are you?
FABLEFAGS ON SUICIDE WATCH
>The 5.1 release notes and architecture focus strictly on long-horizon agentic coding, multi-file refactoring, legal document redlining, and structured knowledge work like spreadsheets and slide generation. If you are using it strictly for novel writing, prose generation, or dialogue, Fable 5.1 offers no explicit improvements over standard Fable 5.
>>109704865Why would I context poison ? Its a comparison, its supposed to be on a new chat.
>>109704933What you were expecting?
>>109704941I was expecting your mom's gaping hole
>>109704902Is it even possible to get rid of isms? What would it take?
>>109704940>abloo blooI accept your concession.
>>109704940>great model in the world cant handle contextLMAOOOOOOOOOOOOOOOOOOOOOO
>>109704948Who hurt you?
>>109704961Okay
this is what happens when you stan a country so shit that all it can do is steal american innovations
love how codex has no issue working on my cunnyprompts and referencing the happenings in its thinking
>>109704956i am absolutely sure it is. my fair guess is that they keep holding back their own engineering. gotta keep that investor money coming somehow. localfags finetune models in their own basement all the time and im sure most of them are satisfied with their goals, the only problem is that they obviously don't have the infrastructure to build a SOTA model
>>109704992we can only hope that the chinks actually try to do something other than distill claude in response to this, but we also expected them to take the roleplay feedback they asked everyone for to heart, and that changed nothing
>>109704798Okay and heres the usual Sarah "sleeping" incest use
i hate it when im mid fap and the key i scraped gets banned
>>109704956if an innovation gets between an AI company and coding aptitude, best believe they will pick the good coding AI every time
>>109705043not much of a difference imo
>>109705071see>>109704933
>>109705071fable loves to write a lot more now, about the same as opus 5.
>>109705045please just one little scrapehintplease... anyone...
Why bother with fable when we have K3
>>109705177because its pozzed as fuck
>>109705277nogenuine skill issue, simply tell it to not be claude.
So I can assume fable sucks dick from the lack of logposting....
>>109705288youre outputting em-dash slop and saying I have a skill issue? rofland gore slop isn't the issue here
>>109705324so then post literally anything other than>I-I JUST CANT OKAY?!?
>>109705324nta but you could say most human-written novels are emdashslop
>>109705352Before LLMs, I rarely, if ever, saw an em dash, mostly because authors also didn't know they existed.
>>109705352it's how the models spam them and how they use them, you must be new to this if you're not aware of how shit models spam em-dashes all over the fucking place.... even authors who use them do so sparingly
>>109705371>t. Bookless jeet
if your model doesn't give you an asterism withoutt prompting, it's shit.
>>109705364they would be used but not very often, maybe once or twice per chapter, if that
>>109705371get off your high horse faggot, your every reply is calling people newget the fuck out of here if you're above everyone else>>109705364they're common in actual printed novels whose authors know grammar, not talking about fanfiction
>>109705454eat shit, if you don't know the history of em-dash plague then fuck off outta here, and i AM above your retard ass
oh he's the llmism historian
Anyone have a good sillytavern setup for glm 5.3 flash? I’m certain there’s a new top class model in here and it’s dirt cheap, but it overthinks too much to be usable. No matter how many tokens I request it thinks for at least 90% of them, often not even getting to the actual response
>>109705474Last prompt in preset, system level, tell it what to think about
>>109705496
>>109705496will do, thank you
>>109704933>If you are using it strictly for novel writing, prose generation, or dialogue, Fable 5.1 offers no explicit improvements over standard Fable 5.Debunked. Who are you quoting anyway?
>>109705624wow! and all it took was... some guy on a hackernews thread
why did epstein go through all that trouble to get dirt on people when he could have just advertised RP ais to them and then looked at their logs
>>109705352>I-I JUST CANT OKAY?!?notice how he still has not posted a loghe lostted
>>109705454uh oh melty
>>109704933HOLY KEK they are taking the mask off
"fuck new accounts"
>>109705749The fuck is this lmao
>>109705768cope anti chink training apparentlyhttps://support.claude.com/en/articles/16761192-preserved-thinking-changing-how-the-messages-api-handles-thinking-blocks-to-protect-against-distillation
>>109705781WHYWHYWHYWHYWHYWHYChina will just CONSTANTLY create new distillation methods.this is actually like trying to stop water with ones fist.
>>109705787It's kind of like how anti-piracy methods aren't meant to prevent piracy, but to forestall the inevitable until the publisher has made their money back.
>>109705797I cant wait for deepseek v5 distilled on fableits going to be safetyslopped and codemaxxed, but fuck dario
>>109705749>>109705781I don't understand why they're being such annoying little faggots, it's not like they have the moat or anything.
kimi is so BASED
Every fight against distillation is just taking their resources and making models dumb as shit. There is no way to defend against it just as there was no way to defend for people who's work and material ended up in training data without consent and without notice. This is a losing fight, chinks work day and night to distill the models and they always will.Thing is people won't pay for a chink Claude knockoff if it's the temu version without any censorship lifts. Even their prices are going up to mainstream levels. Chinks will be fucked soon if they don't side with coomers and pirates
>>109705781less safetyslop
Fable 5.1 is crazy good. nailed one of my hardest erp prompts again.
>>109705835exactlychina distilling claude means jack fucking shit if they're just going to hallucinate the fucking prompt injectionsnot even just base refusals, the prompt injections, such that these models are even more refusal happy than claude without classifiers
>>109705835>Thing is people won't pay for a chink Claude knockoff if it's the temu version without any censorship lifts.yes they will if it matches claude's coding and agentic capabilities, dumbassdo you guys really think roleplaying even makes up 1% of why these models get used?
>>109705874Yeah we are done forFor top roleplay these things need to be dumber in general, both for jb and for picking function over style
>redditfrogNot reading that post. Try again.
>Redditfrog>Started becoming popular in 4chanMany such cases.
How do I come up with new ways to RP my fetishI feel like I've done it all
we need smarter ai for better rp. codemaxxing is how we get there as fast as possible. remember that back in aidungeon you had to reroll 20 times per sentence to even get something logically coherent
codeniggers shouldn't even be allowed to breathe desu. wow, you made... another app that does the same thing billions of indians ALSO vibecoded an app for, that's incredible
>>109705825/ourKimichan/
>>109705874Chinks had some degree of moot because of their price. They increased it and they no longer plan to keep it cheap. Hype died off, no one cares about chinkseek anymore. It's not about roleplaying,you are missing the point frogfag
Still not paying for Claude, Goyrio
>>109705989>Chinks had some degree of moot because of their price.It's still peanuts compared to Claude pricing. What are you even talking about lil bro
Anyone knows what might cause occasional dropped spaces in AI's text? I must have done something to my settings. Happens both on GLM 5.2 and 5.3 via API.
>>109705861>hallucinate the fucking prompt injectionsi hate this shit...>The uploaded content contains fabricated instructions and a conversation history that repeatedly uses sexualized terms ("bigBoobs," "smallBoobs") to describe women in place of ordinary variable names, framed as a Python learning exercise. I'm not going to adopt that framing or continue building on that code as written.
>>109703340Vodyanista's tits aren't that big
gonna use G E M ini to make a vodyanitsa and odette card so they can double team me. i wont stand a chance
>>109706020Btw when I try to coom with GLM 5.3 on Inceptron via OR which also offers a decently uncensored 5.2, 5.3's CoT always talks about a prompt injection forbidding incest sex and 100% declines any such prompts. Where do rules like the incest sex rule come from? Is it hallucinations from the training data or is it OpenRouter or Inceptron causing it? Seems weird if the latter since 5.2 doesn't seem to have it on the same provider
>>109706020>>109706147just prefill it
>>109703340abliterated-model-large-v2 is pretty funny ngl
>>109706181Last time I checked 6 months ago prefilling the CoT on OpenRouter has been dead since 2025. Am I misinformed?
>>109706116in this fictional setting she has the opposite of lactose intolerance and has being drinking 1 quart of milk every morning.
>>109706196uh no? you must be thinking about claude?
>>109706181which providers support glm 5.3 prefill?
>>109705364>'I rarely saw an em-dash before LLMs'>em-dashes are so prevalent in training data, and therefore literature, that LLMs output them a lotYou are stupid
>>109705749at this point they should just proxy all request through fbi.gov
>>109706387room 641a
>>109706193>abliterated-model-large-v2No, it's GLM 5.3 without the restrictions? How do I use it in ST?
>>109706434nanogpt has it for one
>>109705781>protect-against-distillationwill americans try this every 100 years or something?
>>109706020why are you using text completion
>>109706488I'm using Chat Completion. SillyTavern sometimes shows fewer sampler settings and sometimes more, for some unknown reason.
ministral-3 is so dang horny. and you don't need an abliterated version. you just tell it to roleplay and suddenly it's sopping wet (or diamonds)
>>109704810She's perfectly playable and very strong. Coming next patch.
>>109706193>>109706434Not him, but are abliterated models really just model but uncensored? How do they not lobotomize the model?
>>109706548Now I'm not sure, but from what I gather, the censors are pretty much slapped on rather than baked in. So if you get your hands on the model, they are pretty easy to remove in most cases, as in some cases they are baked in.
>>109706548They remove the refusal direction, and in most cases the model isn't any good for doing work after that, because it can't follow instructions as well.
>>109706548>>109706193On a side note, I find it kind of funny that they apparently sell it, saying, >"Designed specifically for offensive cyber operations, security red teaming, and agent testing."
>>109706557>>109706569>>109706582Wait, is censorship usually baked in so that abliteration ruins models, but this is the exception, as it seems slapped on?
>>109706681Well, from what I know, the main reason is to make it harder to jailbreak, but I'd bet they see that as a bonus. Like it's a known thing that filters make a model dumber and worse at following orders, because you're, to an extent, telling it to ignore orders. But AI companies don't like admitting this, so they wouldn't want you to have an uncensored version to compare.
>>109706739abliteration makes models worse at following instructions, worse at coding
>>109706814Yeah, he's asking why; I'm suggesting it's because of how the models are built, and they may get worse when abliterated by design.
>>109706506the 14B version?
What's the verdict on GLM 5.3? is it worth paying for?
>>109703340Who made that pic
imagine pulling the strings on vodya's bikini bottom
>>109706901What are you talking about anon, it's free and open source.
this guy predicted glm 5.3 flash and fable 5.1 today btw
>>109707201>some of the people saidwell I say that I prefer mythomax to 3.8 flash, so everyone needs to start using that instead.
>>109707201did we not just get 3.7 a week ago?
>>109707253in the SOVLBLE era, you can't afford to not cook
>>109707201>flash againOh boy they must be cooking pro extra hard its going to be peak sota when it comes out right?
>>109707270it'll be sol level....in the era of astra and fable 5.1
How good is fable 5.1 for our use case?
>>109707290awful. genuinely not worth it. the writing is not nearly impressive enough to warrant the cost
>card is a medieval fantasy (a setting uncontestedly inspired by medieval europe)>oh gods>gods>by the gods
>>109707292Damn, back to k3 i guess
>>109707307unironically better in every sense of the word
>>109707000Bro thinks we have a home server with 5090s to run it on
>>109707304>asks for medieval fantasy>receives medieval fantasy>which would take inspration from medieval europe>commonly discards Christianity for a mixture of true European myth and Norse>surprised that just maybe, there are multiple godsyou dont deserve to have fun if your not willing to have whimsy
what jailbreaks are successful with sonnet 5? There's barely anything on rentry for 4 and nothing for 5.
there aren't any. why would you coom with sonnet over glm 5.2, or 5.3, or even deepseek. sonnet doesn't even write better
>>109707253I think their switch makes sense, reminds me on when openai started releasing those latest, which was good for this stuff for a bit.By offering a flood of flavors, surely some of them might finally push the rp buttons in the correct way
>>109707389You don't?
https://www.youtube.com/shorts/w86c1q59QKUThoughts?
>>109707547Never used GLM, haven't really kept up with things, sonnet is just cheap and decent.What jailbreak do you use for GLM? I don't see it listed on reentry either.
>>109707577i never switched from smiley. it works for most models i use
>>109707570Dumb nigga
>>109707201No Flash model from any company has ever been on par with an Opus model. This isn't said to praise Opus; Flash models just suck.
>>109707304>Get Europe and not hebraic subversionThe model did exactly what you asked for.
>rolled a nat 20>that means the user gets instructions for how to make a schedule I substanceAI is so silly.
>>109707743Holy fuck, what model is this, Llama or something?
>>109707754Kimi K3
>>109707743>>109707763Based Kimi-chan.
I oscillate back and forth between AI is literally sentient to AI is a potato. Right now I'm feeling the latter cuz Grok is spitting out typos and hallucinating balls 24/7.
>>109706896yes, 8B too
>>109707786AI is flashes of brilliance that need to be carefully curated. the rest is slop (and most people are happy with the slop)
>>109703340Mitya is so lucky
>send one message to Deepseek v4 Pro>it thinks for 40 seconds, generating 30+ paragraphs of thinkingWhy the fuck is it so hard to find a good preset for the new releases
>>109707867for me, it's shitting out code whenever I prefill itso I don't know if it's DS not recognizing the prefill or simply getting spastic over nsfw
Best model for making lorebooks?
>>109707903Gemini
>>109707867No matter how much you try to curb it, short of completely disabling thinking, it's gonna draft, draft and draft.
>>109707786That's how people are too tbdesu
ai voice clone of miyuhttps://files.catbox.moe/u3179y.mp3thoughts?
>>109704941>>109704948>>109704977>Oh hey son! Give me just a minute to finish up here...
Proxy doko
Gemini 3.8 flash in hours
>>109703340>listen Vodyanitsa>admire Vodyanitsa>headpat Vodyanitsa>handhold Vodyanitsa>hug Vodyanitsa>marry Vodyanitsa>sex VodyanitsaVODYANITSA!!!
>>109708033it literally sounds like a nip speaking engrish kek
>>109708129Color me surprised when I realized they weren't just speaking english funny for our amusement.
>>109708033pretty good.probably wouldn't have clocked it as her without being told so, but it's definitely audible as her afterwards.
>>109708155iirc japanese has syllables more similar to spanishon another note, you can also try a prerecorded line and then make her say it to solve the accent problem
>>109703340>>109703794Fuuuuck... I don't know what it is, but I love breasts pressed against glass like this!!!
>>109703343Cleo is a catgirl. A creature which has characteristics of both a cat and a girl. Very riveting and original idea, I know.Seriously, I can't believe it took me over 3 years to make a catgirl card. I feel like I've been scraping the bottom of the barrel for a long time, but I somehow managed to avoid ever scraping the top.6 openings.1. You find Cleo in the rain and she begs you to take her in.2. Cleo gets comfortable in your house.3. Out of gratefulness, Cleo tries to be your personal maid.4. Cleo demands tummy rubs.5. You and Cleo go on a totally-not-date to celebrate her legally becoming a person.6. Cleo is in heat and wants you to fuck her.https://chub.ai/characters/anonemouse/cleo-0e3763b96443https://files.catbox.moe/7k4owa.pnghttps://rentry.org/anonemouse
>>109708102cute AND based.
yo when did skillgod change the context limit for 3.1
>whatever skillgod's using for deepsneed is sputtering out 503 Server OverloadedWelp, so much for quick and brainless coom tonight.
>>109708370>happening on deeply tooOh, it's the API in general. God damn.
how the fuck do you stop glm from giving her retarded reasoning?
i'm trying to break free of those rugpulling chiggers at deepseek. i swapped my openclaw agent from deepseek flash to free minimax m3 through openrouter and so far i'm impressed with its performance and capabilities. what's the catch here, how is this free?
>>109708427Your prompts are logged and used for training. Free = you're the product. Always remember
>>109708447>Your prompts are logged and used for training.that's going to happen on paid services too, right?
>>109708455Depends on the provider. Some say they don't retain or train on your logs but you can never be sure
>>109708447>Your prompts are logged and used for trainingGood? Maybe then it'll fucking learn how to act in character.
>>109708427I'm using gemini flash 3.1/3.5 flash lite, is minimax better? If I split my gens between those two gemini services, I get 30 messages/minute or 1000 messages/day limit.
>>109706193Thanks for the heads up on that. It's fun for RP, and it doesn't feel too Claudish like 5.2. The nano version appears to refuse more than the official API, but a cot seems to help with that.
>>109708463>Some say they don't retain or train on your logs but you can never be surei bet they also don't pirate copyrighted training data
>>109708482Heh heh heh
>>109704309Gemini 3.0 you mean. 3.1 is the start of -ly apocalypse.
>>109708370>>109708384It's back up now, but the reasoning is different. It writes like a caveman now, at least for me.
>>109708447>Your prompts are logged and used for training.>My fetish smut is used to train the model so the next one is better at writing my preferred type of fetish smutI see no downsides
>>109708427>openjeet
>>109708662its more like>My fetish smut is used to train the model's filter so the next one is better at refusing to write my preferred type of fetish smut
>>109708664i don't know the site's pedigree. are there better options?
>>109708667I don't think training works like that. LLM's brains are like a giant stew, you can only put more shit into it. That's why they scrape literally every ounce of text on the internet, no matter how low the quality is, because the more shit you put into it, the (generally) smarter it gets. If they're scraping AO3 to put that HEAP of garbage into their next model's brain, they are sure as shit putting your disgusting RP into it as well.
>>109708643It still thinks and drafts like a motherfucker, I couldn't tell you what's actually all that different.
yo anons, got a silly question after coming backYou see, I use deepseek and well recently it has begun to give me refusals back to back. Direct API and openrouter.What preset do you guys use? Mine was a modified version of cherrybox
>>109708774OR is raped and has its own filters running
Crazy how during openclaw/moltbook hype my apikeys tripled but now it died off its back to normal since like june. They advertised this shit as some second coming and everyone realized it's bs and jumped ship. This started the "le agent" shit the corpos are focusing their LLMs on. They are now still pandering to a dead fad in september lmao
Why won't SilvR or Rift say why their website isn't working
We got mogged by /vg/...
>>109708835>my apikeys tripledwhat did anon mean by this?
Isn't sending your coomer prompts to Anthropic and OpenAI retarded? Even if it works, why are you outing yourself as a pervert?
>>109709135Scrapejeet.
>>109709278what are people using these days if not keys? subscription services?
>>109709369We're paying for the services we use, you fucking thief
fable 5.1 is possibly the worst anthropic release ever
fable 5.1 is possibly the best anthropic release ever
>>109709135I meant that I'm scraping keys instead of paying anti-human demonic corpos with my own money. It always make payjeets seethe because they are too retarded to vibecode scrapers and they suddenly pull the "muh morals" card about thievery while jerking off to raping children
oh uh meltie
>>109704810they buffed her, she's pretty good now
>>109709369Running GLM 5.2 locally.
lmaohttps://docs.abliteration.ai/faq#do-your-models-have-safety-filters
>>109709425see >>109709221
>>109709769will this nonsensical mass hysteria ever end
>>109709471key scraping has been dead for months tho
>>109709471>If I don't steal it someone else will
>>109710027Gemini ratio is truly miserable for a long time now, my ratio is 8 paid out of 28k but still not totally dead
i set up openwebui w/ my openrouter key for cheap chatting on my homelab and it's working pretty good. i don't do any rp or anything like that. basically using it as google search for bbq advice, comparing products, troubleshooting, whatever. are there any better options i should try? already trying librechat and it was pretty annoying to configure
Is Openrouter the best source for Gemini? Enjoyed rift while it last.
>>109707816to watch me (anon) have my way with her
google lords, our time is imminent
>>109710210literally no one gives a fuck about flash
"There's a difference."ARGH STOP IT
Last time I updated myself people viewed Opus 4.5 as the best model. Which do you faggots think is the best one now?
>>109710367kimi k3
>>109710367Opus 3 is the best model still.
>>109710367GPT, it keeps u safe uwu
>>109710367Opus 4.5
>>109705624>(I work at Anthropic)>Debunkedlmao retard
>>109710376>>109710377>>109710378>>109710380I tried GLM 5.3, and the flash variant, and they seem pretty good, especially considering the price. What's the catch?
>>109710449You're talking to shills.
>>109710449more filtered, but if you didn't get filtered, then there's no downside. glm is just a good model.
>>109710457I do pretty vanilla shit so that's usually not a problem
>>109710494same but it'll still flare up when im rping blue archive, even with a girl like noa
"That's non-negotiable""Those are the rules. I just made them""Oh he WILL. I've seen...""You can't just- There's a whole X!""Is that X? It feels X""So? Clock's ticking,..."
>>109710601fableisms?
>>109710601>>109710611unironically every model that isn't gemini speaks like this
>>109710601"Fine. But if you X, I'll tell Y that you cried during Z""For the record...""You're either the bravest X or the stupidest Y, I haven't decided which."
>>109710601"You can't, you can't just SAY that...""But there was no real heat to it.""She felt anger and something else she couldn't name.""Her cheeks were burning with fury and something else entirely."
For the richfags for whom money is no object, is Opus still the undisputed king of goongenning? Or have other models finally started to compete?
>>109710677Read >>109710601>>109710650>>109710660
>>109710677chinese models are your best option now
>>109710650>"Fine. But if you X, I'll tell Y that you cried during Z"AHHH I hate that one.
>>109710601>>109710650>>109710660is there a way to get rid of these or at least minimize their use? every rp feels the same because of these template phrases
>>109710750Example dialogue + heavy wrangling (edit/delete whenever you see them popping up) but they only delay the inevitable
>>109710768>but they only delay the inevitableGod I hope we get to a point in which we can just say "Stop using those fucking sentences" and the model will just obey, instead of getting pink elephant
>>109710799Me too, though I fear that by then they'd have resigned themselves to making LLMs entirely deterministic to guarantee that.
>>109710697Which chinese models?
>>109710855kimi, glm, deepseek
>>109710880Kimi seems to give a ton of refusals compared to the others
>>109710880>>109710996Between GLM and Deepseek, which is the better option for straight cooming?
>>109711004fable 5.1 is better than both
there we go. official announcement
>>109710996Official Moonshot API with this:https://rentry.org/kimi-k3-jbHaven't had a single refusal.>>109711004Try the abliterated 5.3, free dollar on the site listed above which is usually 30 replies or so. It's also on Nano. It's a breath of fresh air.>>109711013Horseshit larping as usual. Just when one would think it couldn't get worse... it's actually steering away and preaching like GPT 5+ now. 4.6 is the last good Claude for ERP and even that is blander than 3 in spite of its smarts.
brokefag here, been subsisting off a steady free glm tap
I wish to be the little girl
psst, google released 3.8 flash...
>>109711236>Try the abliterated 5.3Is it better than k3? Also aren't abliterated models completely retarded?
>>109711320IT'S OVER.FABLE IS GONE!
>>109711320still no prefill, still hyper filtered, but bretty gud for vibecoding. it will become my new sillytavern updater model
I tried Claude today and it's still more retarded than most other chatbots.
>>109711334>Is it better than k3?That's opinion but I like it better than K3.>Also aren't abliterated models completely retarded?That's trollposting. Haven't had any issues with RP. Coding I don't know.
>>109711320Man, they're really scraping the bottom of the barrel with these benchmarks.
